nvfuser::ComputeAtMap::computeConcreteId throws No potential concrete_id's found

[jjsjann123]

hit this assert

" INTERNAL ASSERT FAILED at "/opt/pytorch/nvfuser/csrc/compute_at_map.cpp":1115, please report a bug with repro script to NVFuser at https://github.com/NVIDIA/Fuser/issues. No potential concrete_id's found for disjoint set { iS65{( ceilDiv(( i0 * 4 ), 128) )}; iS81{( ceilDiv(( i2 * i0 ), 128) )}; iS76{( ceilDiv(( i2 * i0 ), 128) )}; iS47{( ceilDiv(( i2 * i0 ), 128) )}; iS89{( ceilDiv(i0, 128) )}; iS85{( ceilDiv(i0, 128) )}; iS53{( ceilDiv(i0, 128) )}; iS41{( ceilDiv(( i2 * i0 ), 128) )}; iS35{( ceilDiv(( i2 * i0 ), 128) )}; iS59{( ceilDiv(( i0 * 4 ), 128) )}; iS71{( ceilDiv(( i0 * 4 ), 128) )} }

Repro test:

TEST_F(PointwiseTest, DomainMapFactory) {
  auto fusion_ptr = std::make_unique<Fusion>();
  auto fusion = fusion_ptr.get();
  FusionGuard fg(fusion);

  TensorView* tv0 = makeContigTensor(1);
  fusion->addInput(tv0);
  TensorView* tv1 = makeContigTensor(2);
  fusion->addInput(tv1);

  auto tv2 = broadcast(tv0, {true, true, false});
  auto tv3 = add(tv2, tv1);
  fusion->addOutput(tv3);

  auto size_val = IrBuilder::create<Val>(4.0, DataType::Int);
  auto one_val = IrBuilder::create<Val>(1, DataType::Int);
  auto tv4 = rand({size_val, one_val, tv0->axis(0)->extent()}, DataType::Float);
  auto tv5 = add(tv2, tv4);
  fusion->addOutput(tv5);

  FusionExecutorCache executor_cache(std::move(fusion_ptr));

  auto options = at::TensorOptions().dtype(at::kFloat).device(at::kCUDA, 0);
  at::Tensor input0 =
            at::empty_strided({25}, {1}, options);
  at::Tensor input1 =
            at::empty_strided({7, 25}, {25, 1}, options);
  auto cg_outputs = executor_cache.runFusionWithInputs({input0, input1});
  // EXPECT_EQ(getVecSizeForPointwise(executor_cache), 4);
  testValidate(fusion, cg_outputs, {input0, input1}, __LINE__, __FILE__);
}

I'm leaning towards this is actually an issue with pointwise scheduler trying to propagate transformation from an illegal reference tv. Opening the issue to track this for now

[jjsjann123]

closing this one, realizing I already opened one in #2514 .